Open-ended generation on SciQ
5.21ECEGrACE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GrACEBackbone=Phi3-3.8B2025.09 | 5.21 | 16.93 | 82.37 | 58.1 | |
| GrACEBackbone=Llama2-7B2025.09 | 5.45 | 14.33 | 87.72 | 57.4 | |
| ApricotBackbone=Phi3-3.8B2025.09 | 7.09 | 18.8 | 76.45 | — | |
| P(True)Backbone=Llama2-7B2025.09 | 7.46 | 24.94 | 49.41 | — | |
| GrACEBackbone=Llama3.1-8B2025.09 | 8.16 | 19.13 | 75.62 | 63.5 | |
| ApricotBackbone=Llama2-7B2025.09 | 8.42 | 16.69 | 82.04 | — | |
| ActCabBackbone=Phi3-3.8B2025.09 | 8.72 | 22.2 | 70.24 | — | |
| Platt ScalingBackbone=Llama3.1-8B2025.09 | 9.03 | 23.37 | 69.46 | — | |
| ActCabBackbone=Llama2-7B2025.09 | 9.34 | 24.51 | 60 | — | |
| Platt ScalingBackbone=Llama2-7B2025.09 | 10.12 | 24.84 | 61.41 | — | |
| Platt ScalingBackbone=Phi3-3.8B2025.09 | 10.83 | 24.57 | 73.69 | — | |
| ActCabBackbone=Llama3.1-8B2025.09 | 10.99 | 21.65 | 69.12 | — | |
| Seq. LikelihoodBackbone=Phi3-3.8B2025.09 | 14.47 | 23.81 | 73.69 | 58.5 | |
| Seq. LikelihoodBackbone=Llama2-7B2025.09 | 16.52 | 26.74 | 61.41 | 57.2 | |
| ApricotBackbone=Llama3.1-8B2025.09 | 17.06 | 24.17 | 69.94 | — | |
| P(True)Backbone=Llama3.1-8B2025.09 | 17.96 | 27.9 | 52.83 | — | |
| VerbalBackbone=Llama3.1-8B2025.09 | 21.11 | 26.2 | 64.09 | — | |
| Seq. LikelihoodBackbone=Llama3.1-8B2025.09 | 24.63 | 27.26 | 69.46 | 62.8 | |
| P(True)Backbone=Phi3-3.8B2025.09 | 31.82 | 30.09 | 79.66 | — | |
| VerbalBackbone=Phi3-3.8B2025.09 | 36.52 | 36.55 | 56.44 | — | |
| VerbalBackbone=Llama2-7B2025.09 | 56.05 | 37.94 | 53.07 | — |